Introduction to Ensemble Learning Featuring Successes in the Netflix Prize Competition Todd Holloway Two Lecture Series for B551 November 20 & 27, 2007 Indiana University Outline • Introduction – Bias and variance problems • The Netflix Prize – Success of ensemble methods in the Netflix Prize • Why Ensemble Methods Work • Algorithms – AdaBoost – BrownBoost – Random forests 1-Slide Intro to Supervised Learning We want to approximate a function, Given examples, Find a function h among a fixed subclass of functions for which the error E(h) is minimal, Independent The distance Variance of the of h from of f predictions Bias and Variance Bias Problem – The hypothesis space made available by a particular classification method does not include sufficient hypotheses Variance Problem – The hypothesis space made available is too large for the training data, and the selected hypothesis may not be accurate on unseen data Bias and Variance Decision Trees • Small trees have high bias. • Large trees have high variance. Why? from Elder, John. From Trees to Forests and Rule Sets - A Unified Overview of Ensemble Methods. 2007. Definition Ensemble Classification Aggregation of predictions of multiple classifiers with the goal of improving accuracy. Teaser: How good are ensemble methods? Let’s look at the Netflix Prize Competition… Began October 2006 • Supervised learning task – Training data is a set of users and ratings (1,2,3,4,5 stars) those users have given to movies. – Construct a classifier that given a user and an unrated movie, correctly classifies that movie as either 1, 2, 3, 4, or 5 stars • $1 million prize for a 10% improvement over Netflix’s current movie recommender/classifier (MSE = 0.9514) Just three weeks after it began, at least 40 teams had bested the Netflix classifier. Top teams showed about 5% improvement. However, improvement slowed… from http://www.research.att.com/~volinsky/netflix/ Today, the top team has posted a 8.5% improvement. Ensemble methods are the best performers… Rookies “Thanks to Paul Harrison's collaboration, a simple mix of our solutions improved our result from 6.31 to 6.75” Arek Paterek “My approach is to combine the results of many methods (also two- way interactions between them) using linear regression on the test set. The best method in my ensemble is regularized SVD with biases, post processed with kernel ridge regression” http://rainbow.mimuw.edu.pl/~ap/ap_kdd.pdf U of Toronto “When the predictions of multiple RBM models and multiple SVD models are linearly combined, we achieve an error rate that is well over 6% better than the score of Netflix’s own system.” http://www.cs.toronto.edu/~rsalakhu/papers/rbmcf.pdf Gravity home.mit.bme.hu/~gtakacs/download/gravity.pdf When Gravity and Dinosaurs Unite “Our common team blends the result of team Gravity and team Dinosaur Planet.” Might have guessed from the name… BellKor / KorBell And, yes, the top team which is from AT&T… “Our final solution (RMSE=0.8712) consists of blending 107 individual results. “ Some Intuitions on Why Ensemble Methods Work… Intuitions • Utility of combining diverse, independent opinions in human decision-making – Protective Mechanism (e.g. stock portfolio diversity) • Violation of Ockham’s Razor – Identifying the best model requires identifying the proper "model complexity" See Domingos, P. Occam’s two razors: the sharp and the blunt. KDD. 1998. Intuitions Majority vote Suppose we have 5 completely independent classifiers… – If accuracy is 70% for each • 10 (.7^3)(.3^2)+5(.7^4)(.3)+(.7^5) • 83.7% majority vote accuracy – 101 such classifiers • 99.9% majority vote accuracy Strategies Boosting – Make examples currently misclassified more important (or less, in some cases) Bagging – Use different samples or attributes of the examples to generate diverse classifiers Boosting Make examples currently misclassified more important (or less, if lots of noise). Then combine the hypotheses given… Types • AdaBoost • BrownBoost AdaBoost Algorithm 1. Initialize Weights 2. Construct a classifier. Compute the error. 3. Update the weights, and repeat step 2. … 4. Finally, sum hypotheses… Classifications (colors) and Weights (size) after 1 iteration Of AdaBoost 20 iterations 3 iterations from Elder, John. From Trees to Forests and Rule Sets - A Unified Overview of Ensemble Methods. 2007. AdaBoost • Advantages – Very little code – Reduces variance • Disadvantages – Sensitive to noise and outliers. Why? BrownBoost • Reduce the weight given to misclassified example • Good (only) for very noisy data. Bagging (Constructing for Diversity) 1. Use random samples of the examples to construct the classifiers 2. Use random attribute sets to construct the classifiers • Random Decision Forests Leo Breiman Random forests • At every level, choose a random subset of the attributes (not examples) and choose the best split among those attributes • Doesn’t overfit Random forests • Let the number of training cases be M, and the number of variables in the classifier be N. For each tree, 1. Choose a training set by choosing N times with replacement from all N available training cases. 2. For each node, randomly choose n variables on which to base the decision at that node. Calculate the best split based on these. Breiman, Leo (2001). "Random Forests". Machine Learning 45 (1), 5-32 Questions / Comments? Sources • David Mease. Statistical Aspects of Data Mining. Lecture. http://video.google.com/videoplay?docid=- 4669216290304603251&q=stats+202+engEDU&total=13&start=0&num=10&so=0&type=search&plindex=8 • Dietterich, T. G. Ensemble Learning. In The Handbook of Brain Theory and Neural Networks, Second edition, (M.A. Arbib, Ed.), Cambridge, MA: The MIT Press, 2002. http://www.cs.orst.edu/~tgd/publications/hbtnn-ensemble-learning.ps.gz • Elder, John and Seni Giovanni. From Trees to Forests and Rule Sets - A Unified Overview of Ensemble Methods. KDD 2007 http://Tutorial. videolectures.net/kdd07_elder_ftfr/ • Netflix Prize. http://www.netflixprize.com/ • Christopher M. Bishop. Neural Networks for Pattern Recognition. Oxford University Press. 1995..